‘Stealth corrections’: when journals quietly fix papers

Last March, René Aquarius noticed some overlapping patterns in a figure about a 2016 study on the blood-brain barrier. So he took to PubPeer, an online site where scientists often discuss papers, to raise his concerns. 

An author of the  study published in Neuroscience Letters responded saying they are checking the original data to figure out the problem. A month later, when Aquarius, a postdoctoral researcher at Radboud University Medical Center in Nijmegen, Netherlands, revisited the paper, the figure had been replaced without any note that the publisher had fixed the issue. 

Aquarius once again took to PubPeer to express his concerns. “I don’t see any notification when looking at the landing site for the paper: no erratum, corrigendum or a simple log-entry that something has been changed,” he wrote, noting that he had informed Elsevier, the journal’s publisher about the issue. In July, the journal issued a corrigendum for the paper. 

“I was quite a bit upset about it,” Aquarius told Retraction Watch. “It takes away one of the key elements for any reader to be critical, namely that you know what has happened.”

What Aquarius uncovered wasn’t an isolated case. He and his fellow research-integrity sleuths, who police various issues in scholarly literature alongside their day jobs, found 130 more cases of what they dub stealth corrections, where journals fix papers without acknowledging that they have done so. They outline their findings in a paper published as a preprint on arXiv on September 10. 

“The fact that they are interested in correcting the record is commendable but the way in which it’s being done is clearly suboptimal and suggests in some cases it’s self serving rather than serving science as a whole,” said Tom Heyman, a statistics and methodology researcher at Leiden University in the Netherlands, who wasn’t involved with the new study. 

Aquarius and colleagues found the cases of stealth corrections by revisiting papers flagged on PubPeer and via social media accounts of sleuths. Aquarius said  the scale of the problem is unclear.  One would need to have access to different versions of papers before and after changes are made to spot any tweaks. 

Of the 131 papers in the study sample, 76 were published by the Journal of the Balkan Union of Oncology. None has an associated digital object identifier but all have PubMed IDs. We reached out to the journal and its publisher, Bakis Productions, for comment but have not heard back. 

“At MDPI, we broadly agree with the overall sentiment of the article,” a spokesperson for the publisher said. “To this end, we have fully updated our correction process to increase the overall transparency of our post-publication updates, using ‘minor correction’ as a way of publicly announcing non-scientific changes to a manuscript.”

But removing articles published in a special issue from the digital special issue website shouldn’t be considered a stealth correction as the paper suggests, the MDPI spokesperson added in a statement provided to Retraction Watch. “[W]e are very transparent [about] our process, and as this change does not impact the scientific record, we do not accept that a public correction notice is required.”

“Articles that have been published shall remain extant, exact and unaltered as far as is possible. However, circumstances may arise where an article needs to be corrected, retracted or even removed,” a spokesperson for Elsevier added. “We do not correct articles without a formal notice and our policy outlines this. We will investigate the journal papers that have been brought to our attention.”

Although Aquarius and his team are pleased journals are fixing problems, they argue doing so silently is not the right solution. “You’re never sure who might have read the paper in the meantime and is not aware of the problem that was there and has been addressed,” Aquarius said. 

Another issue journals have been silently fixing are fingerprints left by generative AI tools like ChatGPT when they are used to write manuscripts, said study co-author Guillaume Cabanac, a computer scientist at the University of Toulouse in France. Cabanac has developed a tool that finds mentions of odd turns of phrases that are indicative of fabricated research papers. 

“We would like the authors to be accountable,” Cabanac said. “If they use any technology and don’t disclose it, that’s a problem,” he added. Journals should be asking authors to what extent they used ChatGPT and if they checked the tools’ outputs, Cabanac says. 

Other corrections being fixed quietly include paper titles, authors names, ethical disclosures, images, abstracts, affiliations, among others, Cabanac said. For him, one way to implement what he calls “enforced transparency” might be in the form of blockchain technology. “You cannot cook the books with the blockchain technology,” he said. “Once the information is stored, you cannot alter it without the public knowing about it.” 

Julia Rohrer, a psychologist at Leipzig University in Germany who wasn’t involved with the study, called stealth corrections “very troubling.” Rohrer, who previously ran the Loss of Confidence project — which provided a platform for scientists to document instances when they lose faith in their previous findings — said removing figures, authors, and markers of ChatGPT use is particularly concerning. 

“It feels like we’re not at the level of transparency that we should be as a science as a whole,” Heyman, who co-authored a 2020 study on correction notices issued by psychology journals, added. 

Heyman notes that different types of issues are being corrected publicly and stealthily. For instance, the new study found that many stealth corrections are of the content of papers while Heyman’s 2020 study on publicly issued corrections identified changes in metadata such as information about funders. 

“If you look at the content changes that were stealthily corrected, they seem often to involve really impactful changes,” Heyman said. “Whereas, if you look at the public correction notices, they regularly mention that there were changes but they were minor [and] they didn’t affect the conclusions.”

In some cases, corrections may be framed as making minor changes, Heyman said, but they may affect the conclusions. “So you also have a lack of transparency at that level,” he said. “Yes, there the correction is public but then the way the correction is being described is maybe not as transparent as it could have been.”

Journals don’t typically ask peer reviewers to assess corrections, corrigenda and errata, but doing so might make it harder for authors to falsely claim tweaks don’t affect the study’s conclusions, Heyman said. In addition to software companies sending researchers notifications when they cite retracted papers, Rohrer said similar alerts should also exist for lengthy corrections. 

As for corrections that are more benign, Rohrer said she is fine with those being corrected stealthily. Recently, Rohrer herself was involved with a stealth correction in one of her papers where she forgot to include her middle initial, which, she noted, raises particular problems in citation management software. 

“I think it’s perfectly fine for these instances,” Rohrer said. “Given that people’s attention is a limited resource, I think what we additionally need besides transparently tracking all changes and a complete version history, is some clear guidelines for what needs to be flagged and what does not.”

Like Retraction Watch? You can make a tax-deductible contribution to support our work, follow us on Twitter, like us on Facebook, add us to your RSS reader, or subscribe to our daily digest. If you find a retraction that’s not in our database, you can let us know here. For comments or feedback, email us at [email protected].